Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Curvature Generation in Curved Spaces for Few-Shot Learning
Zhi Gao, Yuwei Wu, Yunde Jia, Mehrtash Harandi
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add HAN with fixed c=1 to prototype averaging (ablation) | Curvature generation FSL · mini-ImageNet 1-shot inductive, BigResNet12 | accuracy: 59.05 -> 62.60 | Improved |
| add CCG (class-level curvatures) on top of HAN (ablation) | Curvature generation FSL · mini-ImageNet 1-shot inductive, BigResNet12 | accuracy: 62.60 -> 67.02 | Improved |
| generate class-level curvatures instead of a single curvature (ablation) | Curvature generation FSL · mini-ImageNet 1-shot inductive, BigResNet12 | accuracy: 58.97 -> 59.47 | Improved |
| generate curvature (CCG) instead of fixed c=0.01 with HAN (ablation) | Curvature generation FSL · mini-ImageNet 1-shot inductive, BigResNet12 | accuracy: 64.17 -> 67.02 | Improved |
| fixed curvature c=0.01 instead of c=1 without CCG and HAN (ablation) | Prototype averaging · mini-ImageNet 1-shot inductive, BigResNet12 | accuracy: 59.05 -> 50.92 | Hurt |
| fixed curvature c=0.01 instead of c=1 with HAN (ablation) | HAN · mini-ImageNet 1-shot inductive, BigResNet12 (5-shot) | accuracy: 1-shot 62.60 -> 64.17; 5-shot 79.25 -> 76.49 | Mixed |
| increase m from 1 to 4 collected samples (ablation) | HAN · mini-ImageNet 1-shot inductive | accuracy: 65.03 -> 67.02 | Improved |
| increase m from 15 to 60 collected samples (ablation) | HAN · mini-ImageNet 1-shot transductive | accuracy: 65.88 -> 71.79 | Improved |
| increase m from 60 to 75 (ablation) | HAN · mini-ImageNet 1-shot transductive | accuracy: 71.79 -> 71.11 | No clear effect |
| full method vs model-capacity-matched variant (ablation) | Curvature generation FSL · mini-ImageNet 1-shot transductive, BigResNet12 | accuracy: 68.25 -> 71.79 | Improved |
| generate curvature from second-order statistics instead of concatenation (ablation) | CCG · mini-ImageNet 5-shot transductive | accuracy: 82.04 -> 83.00 | Improved |
| generate curvature from second-order statistics instead of mean (ablation) | CCG · mini-ImageNet 5-shot transductive | accuracy: 80.73 -> 83.00 | Improved |
Closest papers by meaning
- Adaptive Poincaré Point to Set Distance for Few-Shot Classification (AAAI 2022 · similarity 0.90)
- Learning Mixed-Curvature Representations in Product Spaces (ICLR 2019 · similarity 0.89)
- Few-Shot Learning With Embedded Class Models and Shot-Free Meta Training (ICCV 2019 · similarity 0.88)
- Exploring Complementary Strengths of Invariant and Equivariant Representations for Few-Shot Learning (CVPR 2021 · similarity 0.86)
- Dual-Geometry Graph Network: Unifying Local and Global Priors for Few-Shot Learning (AAAI 2026 · similarity 0.86)
- Robust Hyperbolic Learning with Curvature-Aware Optimization (NeurIPS 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
